The European Union has started enforcing new parts of its AI Act. The rules took effect on August 3, 2026, and focus on transparency and oversight of artificial intelligence systems.
The European Commission's AI Office is leading the effort. It is working with national authorities across EU member states to apply the new requirements.
One of the main changes affects chatbots and virtual assistants. These systems must now clearly tell users they are talking to AI rather than a human being.
The goal is to help people make informed choices when they interact with automated tools. Users should not mistake AI systems for real people.
Henna Virkkunen, Executive Vice-President for Tech Sovereignty, Security and Democracy, spoke about the changes. She said AI can bring benefits to people and businesses, but risks grow as systems become more capable.
She added that the AI Act creates a clear, risk-based framework. This is meant to give businesses legal certainty while protecting the public.
New Transparency Rules for Businesses
AI-generated media now faces stricter labelling rules. Images, videos, and audio created or altered by AI, often called deepfakes, must be clearly marked.
This content must also include machine-readable identifiers. These tags help platforms and regulators detect AI-generated material more easily.
To help companies comply, the Commission published a list of organizations that signed its voluntary Code of Practice. More than 180 groups have joined so far.
Oversight and Future Deadlines
The AI Office also gained new power to supervise general-purpose AI models. These are the foundation systems behind many chatbots, search tools, and AI agents.
Providers of these models must keep technical documentation and share information with regulators. They must also publish summaries of the data used to train their systems.
Models seen as having systemic risk face extra rules. Providers must address risks tied to cyberattacks, manipulation, loss of human control, and chemical or biological threats.
Enforcement duties are split between several bodies. The AI Office handles large platforms and search engines, national authorities cover other systems, and the European Data Protection Supervisor oversees EU institutions.
A new Scientific Panel of 60 independent experts will support this work. Professor Alessandro Abate of Oxford University was named Lead Scientific Adviser.
The Commission also launched new reporting tools. A Complaint Tool lets people report suspected breaches, and separate whistleblower channels are open to employees and other individuals.
Looking ahead, some deadlines have shifted. Rules for high-risk AI systems are now delayed until December 2, 2027.
High-risk AI built into regulated products will not apply until August 2, 2028. The Commission adjusted these dates through what it calls the AI Omnibus package.
One rule keeps its original deadline. A ban on AI systems used to create non-consensual sexual images and child sexual abuse material takes effect on December 2, 2026.
